We have developed databases and software useful for comparative analysis of protein three-dimensional structure. These tools are distributed freely to biologists and developers of biotechnology software. MMDB (Molecular Modeling DataBase) is the 3D-structure component of the Entrez molecular biology retrieval system. MMDB is an ASN. 1 database where all data items describing macromolecular structure are validated and explicitly listed, so that application software need not contain the complex logic required to retrieve this information from text formats such as PDB files. Work has concentrated on addition of accurate taxonomy assignments for macromolecular structures within MMDB, creation of new message and data types for transmission of structure-structure alignment data to local viewers, and on construction of an automated monthly update and indexing system, Pubstruct. CN3D ("see in three dimensions") is a multi-structure visualization program distributed as part to the Entrez client software and in a stand-alone version lauchable via the MIME protocol in World-Wide-Web Entrez. The software differs from other public domain viewers in supporting display of multiple aligned structures from Entrez's "structure neighbor" database, and in supporting simultaneous highlighting/picking of multiple sequence and multiple structure alignments. Other features added this year are on-the-fly alignment of the sequences of homologs, so that an Entrez user may easily map conserved sequence features onto the know 3D structure. These software features are intended to facilitate molecular biologist's identification of important structure-function relationships within protein families. Work this year has concentrated on improvements to CN3D. The software has been modified to use an industry-standard 3D graphics library, Open GL, which provides much better quality molecular graphics rendering. We have also added core-structure alignment editing and threading tools to the sequence display windows, to support curation of CDD (a Conserved Domain Database). Work is in progress to revise and simplify the data structures underlying CN3D, so that further improvments in graphcis presentation, specific to describing conserved features in protein families, may be added to future versions of CN3D.